Inductive learning of attribute path values in typed-unification grammars

Abstract. This work is aiming to show that inductive logic programming (ILP) is a suitable tool to learn linguistic phenomena in typed-unication grammars, a class of attribute-value grammars increasingly used in natural language processing (NLP). We present a strategy for generating hypothesis spaces for either generali-sation or specialisation of attribute-path values inside type denitions. This strategy is the core of a prototype module, extending the LIGHT system for parsing with typed-unication grammars.